Publication (Phd Thesis)

Control algorithms for energy optimization in multimedia hand-held devices

Tang, Qiong
The micro-electronics industry has been boosting the capabilities of multimedia mobile devices, but the battery, which is the only power source of most mobile devices, is experiencing relatively slow development. Therefore, determining how to optimize the energy consumption of mobile devices under a predefined performance requirement has become a critical issue. Besides, according to a recent report, tablet and smartphone video consumption grew 35% in the year 2014 and has grown 170% since 2013. Actually, mobile video playback has experienced a significant growth of 2084% from 2011 to 2015. As some of the most energy-consuming tasks, encoding, decoding and presentation of video sequences are among the main subjects of research on power management in multimedia systems. In addition, every new video standard also tends to increase the energy requirement of video tasks with respect to the previous standards. This dissertation presents a solution based on control algorithms for power regulation under the limited battery capacities of multimedia hand-held devices while executing a decoder application and maintaining a reasonable quality of user experience. A control system, which includes a real-time closed-loop control subsystem and a power-control governor, is proposed and it has been implemented in the operating system of a low-cost development board. Instead of using any specific power sensor, a power estimator based on monitored system events of multimedia mobile devices is proposed as the feedback signal in the closed-loop subsystem. The power estimator periodically obtains significant-events count values and calculates powerconsumption estimations through mathematical models. This power estimator has been implemented in a Linux kernel and evaluated while running a video decoder application on an embedded development platform. Afterwards, prior to the implementation of the real-time control system, off-line estimation data are used to get a system model, which enables the application of classic control-theory methods to analyze and design different controllers. The simulation results show that integral controllers keep the system stability and achieve null average steady-state error with short settling times, even in the presence of estimation noise or disturbance. From these promising simulation results, the controllers have been implemented in the development board and the real results match simulation results. The control system is able to regulate the power consumption and the battery discharge rate in the presence of fluctuations in the decoder power-consumption demand, which presents good results to guarantee a certain battery lifetime.
Research areas:
Type of Publication:
Phd Thesis
Type of Publication:
PhD Thesis